Many-to-Many Voice Conversion based Feature Disentanglement using Variational Autoencoder

TL;DR

This paper introduces a novel variational autoencoder-based method for many-to-many voice conversion that effectively disentangles speaker identity and linguistic content, enabling conversion across multiple speakers including unseen ones, with improved naturalness and similarity.

Contribution

The proposed feature disentanglement approach allows a single autoencoder to perform many-to-many voice conversion and handle unseen speakers, advancing the state-of-the-art in VC.

Findings

Competitive performance in naturalness and speaker similarity.

Effective disentanglement of speaker identity and content.

Handles unseen target speakers naturally.

Abstract

Voice conversion is a challenging task which transforms the voice characteristics of a source speaker to a target speaker without changing linguistic content. Recently, there have been many works on many-to-many Voice Conversion (VC) based on Variational Autoencoder (VAEs) achieving good results, however, these methods lack the ability to disentangle speaker identity and linguistic content to achieve good performance on unseen speaker scenarios. In this paper, we propose a new method based on feature disentanglement to tackle many to many voice conversion. The method has the capability to disentangle speaker identity and linguistic content from utterances, it can convert from many source speakers to many target speakers with a single autoencoder network. Moreover, it naturally deals with the unseen target speaker scenarios. We perform both objective and subjective evaluations to show the competitive performance of our proposed method compared with other state-of-the-art models in terms of naturalness and target speaker similarity.